Pith. sign in

REVIEW 3 major objections 5 minor 42 references

IDENT Review: Recent Advances in Identification of Differential Equations from Noisy Data

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This review establishes that identifying a differential equation from a single noisy observation reduces to solving a linear system for a sparse coefficient vector, and that three ingredients—denoising, sparsity, and model…

desk verdict A useful, clearly-written tour of the authors' own IDENT pipeline, but too one-sided and overclaiming in its conclusion to stand as a balanced field review. read the letter →

arxiv 2506.07604 v1 pith:MHAYNBF2 submitted 2025-06-09 math.NA cs.NA

Roy Y. He , Hao Liu , Wenjing Liao , Sung Ha Kang This is my paper
classification math.NAcs.NA MSC 65L0965M3265M0662J07
keywords differentialequationidentificationsingletrajectorynoisydatasparseregressionmodelselectionweakformulationsubspacepursuitvariablecoefficients
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This review article is about a specific inverse problem: given one noisy trajectory of a solution, recover the differential equation that generated it. The authors' starting point is that if the right-hand side is assumed to be a linear combination of user-selected feature terms (monomials and derivatives), then the problem becomes a linear system $b = F c$ in which the support of $c$ gives the form of the equation and the values give the coefficients. Surveying a family of methods built on this reduction, the paper identifies three ingredients that jointly determine accuracy: proper denoising of the feature system, careful support recovery through sparsity combined with model selection, and refinements such as high-dynamic-region or narrow fits for coefficient values. The authors argue that with these ingredients, constant- and variable-coefficient PDEs—including Burgers, KdV, Kuramoto–Sivashinsky, reaction–diffusion, and nonlinear Schrödinger—can be identified from a single noisy trajectory without any training.

What carries the argument

The central object is the discrete feature system $b = F c$: $b$ is a numerical approximation of the time derivative $u_t$, each column of $F$ approximates one dictionary feature such as $1$, $u$, $u_x$, $u^2$, $u u_x$, and $c$ is the unknown coefficient vector whose support is the equation's form. The mechanisms that carry the argument are the sparsity-plus-model-selection pipeline for finding that support, and the error analysis that compares finite-difference, successively denoised differentiation, and weak-form constructions of $F$ and $b$. In the weak form, derivatives are moved onto smooth test functions by integration by parts, which removes the dependence of the discretization error on the derivative order and provides a built-in denoising effect through convolution.

What would settle it

Take Burgers' equation with a deliberately added term not in the dictionary, e.g. $u_t = -u u_x + 0.1 u^3$, run the pipeline with a dictionary that omits $u^3$, and check whether the selected model still has low time-evolution error or reduction-in-residual score. If a wrong support with correlated terms passes all model-selection criteria, the claim that the three themes are sufficient for correct identification is refuted; a correct method should flag the missing term through a large residual.

Watch

Extended reading notes

Core claim

The paper's central claim is that identifying a differential equation from data reduces to finding the coefficient vector $c$ in the discrete linear system $b = F c$, where $b$ approximates $u_t$ and the columns of $F$ approximate the candidate feature terms; the nonzero entries of $c$ (its support) name the terms in the equation and their values give the coefficients. It further claims that this identification can be made robust from a single noisy trajectory by combining three common themes: (i) building the feature system with denoising—least-squares moving average for finite-difference features, successive denoised differentiation for higher derivatives, or a weak integral form whose test functions act as low-pass filters; (ii) using sparsity (LASSO or subspace pursuit) only to generate candidate supports, then selecting among them with mathematically motivated criteria such as time-evolution error, multi-shooting time-evolution error, cross-validation, or reduction in residual; and (iii) refining coefficient values, for example by fitting only in high-dynamic regions, trimming negligible features, and normalizing columns of the feature matrix. A key supporting claim is that the weak-form error bound is independent of the order of derivatives in the underlying PDE, whereas finite-difference error bounds degrade with derivative order and noise.

Load-bearing premise

The load-bearing premise is that the true equation's right-hand side is exactly a linear combination of a finite, user-specified dictionary of feature terms; if the true dynamics contain a term outside the dictionary, the linear system cannot represent the equation and the whole approach collapses.

Editorial extensions

If this is right

  • If the three-ingredient recipe is correct, single noisy trajectories are enough for identifying a wide class of PDEs, without needing multiple realizations or training data.
  • Higher-order and higher-noise identification becomes feasible by switching from finite-difference features to weak-form features, since the error no longer scales with the derivative order in the same way.
  • Variable-coefficient equations can be handled by expanding coefficients in basis functions and enforcing group sparsity, with model selection via residual reduction.
  • The noise-to-signal ratio defined from the feature matrix explains why two equations at the same percentage noise can have very different identification difficulty.
  • The model-selection criteria are generic enough to be paired with any sparse regression method, so improvements in sparse solvers could directly improve identification.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The dictionary itself is the real bottleneck; an adaptive or learned dictionary that grows candidate terms in response to residual structure would be a natural extension, though such adaptivity is not addressed in the reviewed methods.
  • The weak-form error bound suggests that the support size of the test function is a tunable denoising parameter, so choosing it by validation could further improve identification beyond the fixed choices reported.
  • The same pipeline should apply to identifying fractional derivatives or delay terms only if those terms are included in the dictionary, since the linear-system representation does not itself restrict the form of the features.
  • The paper's emphasis on interpretable model selection could be carried over to experimental video data, where the main anticipated difficulty is data acquisition rather than algorithmic identification.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. This manuscript is a review of a line of work on identifying differential equations from a single noisy trajectory, centered on the IDENT family of methods (IDENT, Robust IDENT, Weak IDENT, Fourier IDENT, GP-IDENT, CaSLR, WG-IDENT). The paper frames the identification problem as a linear system b=Fc built from a dictionary of candidate feature terms, and identifies three common themes: denoising the feature system, support identification via sparsity and model selection, and coefficient recovery refinements. It presents error bounds for finite-difference, successively denoised differentiation, and weak-form formulations, summarizes identifiability results from a cited paper, introduces a new noise-to-signal ratio, and reports selected numerical results from the original papers. The conclusion claims that the reviewed methods robustly identify the underlying equation from one noisy realization without requiring training.

Significance. As a survey of a specific research program, the manuscript is useful: it collects in one place the formulations, denoising strategies, model-selection criteria, and error estimates of the IDENT-family methods, and it makes the new NSR formula explicit in Eq. (24). The paper does not provide new numerical experiments or reproducible code, but the quoted error bounds and identifiability theorems come from peer-reviewed sources and the review is generally faithful to them. If the central claims are properly qualified, the survey could help practitioners navigate the denoising/sparsity/model-selection design space. However, the paper's evidence is almost entirely self-citational, and its own identifiability analysis in Section 3.2 substantially limits the unqualified conclusion in Section 8; those limitations must be addressed before the review can be considered balanced.

major comments (3)
  1. [Section 8; Section 3.2] The concluding remarks claim that the reviewed methods robustly identify the underlying differential equation from one single noisy trajectory and provide 'optimally fitted results for datasets of any size.' This drops the identifiability qualification established earlier in Section 3.2: Eqs. (13)-(14) and Theorem 3.2 of [18] show that unique recovery from a single trajectory requires the solution to contain sufficiently many Fourier modes, and for elliptic operators every trajectory lies within epsilon of a space of dimension O(|log epsilon|^2), making recovery intrinsically ill-conditioned. None of the methods in Sections 4-6 check this condition, and the model-selection criteria TEE, MTEE, CEE, and RR will return a best candidate even when the feature matrix is rank-deficient or coherent. The abstract and conclusion should either be qualified with this information condition or the paper should propose a data-dependent check; Section 7.2.1 shows that such conditions can be stated explicitly.
  2. [Section 2, Eq. (2)] The dictionary completeness assumption, that the true equation is a linear combination of the prescribed feature terms, is load-bearing and is never relaxed. If a term outside the dictionary is active, the linear system b=Fc is misspecified and the subsequent denoising, sparsity, and model-selection steps cannot recover the missing term; model selection will instead choose the least bad dictionary model. Although the assumption is stated at the outset and repeated in Section 8, the paper does not discuss its consequences, how it might be tested, or what happens under model mismatch. A review that claims robust identification should make this limitation as prominent as the claimed successes.
  3. [Sections 4 and 6; Table 2] The evidence for the paper's main thesis, that the three common themes yield accurate identification, consists almost entirely of the authors' own IDENT-family papers. Table 2 summarizes only these methods plus SINDy, and the comparative numerical tables are quoted from the original papers rather than produced by an independent benchmark on a common data set. Because the same group developed the methods and the model-selection criteria, the review cannot rule out that the observed success of the three themes is specific to this test suite. The manuscript should either include a small independent comparison on shared data or explicitly label the survey as covering the authors' research program and state the absence of independent validation.
minor comments (5)
  1. [Throughout] There are numerous typographical errors: 'spacial' should be 'spatial', 'nonliner' should be 'nonlinear' (Sections 1, 6, 8), 'Comparied' should be 'Compared' (Section 6), and 'Dentoe' should be 'Denote' (Section 5.2).
  2. [Sections 4.2.1 and 4.2.3] The words 'psudinverse' and 'submatirx' should be corrected to 'pseudoinverse' and 'submatrix', respectively.
  3. [Section 6, Table 5] The equation name 'Korteweg-de Vires' should be 'Korteweg-de Vries' in both the table and the surrounding text.
  4. [Section 3.1, Eq. (9)] The notation uses delta t for the simulation time step and Delta t for the sampling time step; the formula should state explicitly that these are independent discretization parameters, as the current text may confuse the two.
  5. [Section 6, Eq. (24)] The new NSR is defined using the true coefficient vector c and the true support, so it is not a data-only diagnostic in practice; the paper should note that it is an a-posteriori explanatory quantity for experiments rather than a model-selection criterion.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: this is a review of the authors' own methods with explicit assumptions and external comparisons; no load-bearing step reduces to its own input.

full rationale

The paper is an expository review of a line of work from a single research group, but none of its derivations or conclusions rely on assuming what it claims to establish. The central reduction from a PDE to a linear system is an openly stated modeling assumption: 'we assume that the governing differential equation of u(x, t) is a linear combination of various linear and nonlinear differential terms' (Eq. (2)), and the identification problem is then formulated as solving b = Fc. This is not circular; it is the starting premise of the reviewed methods. The three 'common themes' (denoising, support identification, coefficient recovery) are supported by numerical experiments in the reviewed papers, including comparisons against external methods such as PDE-FIND [33], SINDy [7, 32], and Weak SINDy [26, 27], so the central claims have independent empirical content. The identifiability results in Section 3.2 are cited from [18], which shares an author with the present paper, but they are presented as external theorems with explicit conditions (Fourier-mode richness, elliptic versus hyperbolic operators, Kolmogorov n-width bounds) and are not used to force the methodological choices; they are context and a caution about informative data. The model-selection criteria TEE, MTEE, CEE, and RR are heuristics selected by the authors, not predictions derived from fitted parameters, and the NSR metric in Eq. (24) is explicitly said to be 'derived from [21, Theorem 1]' and is used as a diagnostic rather than as a fitted prediction. Self-citation is frequent, but per the supplied rules it is not circularity when, as here, the cited results are theorems or experiments with stated assumptions and external comparisons. No specific equation is shown to equal its input by construction, and no fitted parameter is renamed as a prediction. The review has no significant circularity.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim of the review rests on the dictionary assumption and the identifiability conditions from the authors' prior work. The hyperparameters are inherited from the reviewed methods and are chosen heuristically.

free parameters (5)
  • Sparsity level k = various
    The number of nonzero terms in the identified PDE is chosen by model selection criteria (TEE, MTEE, CEE, RR).
  • LASSO penalty lambda = not specified
    Balancing parameter in the l1-regularized problem (Eq. 15); its value affects support recovery.
  • Threshold rho = 0.05 (trimming), 0.015 (RR)
    Used to trim small features and to decide when residual reduction is negligible; chosen by hand in the reviewed works.
  • NRR = 5
    Window size for the reduction-in-residual criterion (Eq. 20).
  • Test function support and smoothness (mx, mt, px, pt) = varies
    Controls the denoising effect and integration accuracy in the weak-form methods; chosen by hand in [26].
assumptions (5)
  • domain assumption The underlying PDE is a linear combination of a finite dictionary of feature terms (Eq. 2).
    The entire linear-system formulation depends on the true equation being in the dictionary; stated in Section 2.
  • domain assumption Observed data are the true solution plus independent Gaussian noise (Eq. 1).
    Error analyses in Section 3 assume this noise model.
  • domain assumption For weak-form methods, test functions vanish on the boundary and the support size h is adjustable.
    Integration by parts in Eq. (7) requires boundary conditions on the test function; see Section 3.0.3.
  • domain assumption Identifiability requires the trajectory to contain sufficiently rich dynamics (e.g., enough Fourier modes).
    Theorem results from [18] state identifiability conditions that are assumed to hold for the success of the methods; discussed in Section 3.2.
  • standard math Numerical schemes are consistent, stable, and convergent, so TEE/MTEE select the true PDE.
    The model selection rationale in Section 4.2.1 relies on Lax-Richtmyer equivalence for the candidate equations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of IDENT Review: Recent Advances in Identification of Differential Equations from Noisy Data." pith.science (2026). https://pith.science/paper/MHAYNBF2

@misc{pith2026250607604,
  author       = {Pith},
  title        = {Pith review of: IDENT Review: Recent Advances in Identification of Differential Equations from Noisy Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MHAYNBF2}},
  note         = {Machine review of arXiv:2506.07604}
}
read the original abstract

Differential equations and numerical methods are extensively used to model various real-world phenomena in science and engineering. With modern developments, we aim to find the underlying differential equation from a single observation of time-dependent data. If we assume that the differential equation is a linear combination of various linear and nonlinear differential terms, then the identification problem can be formulated as solving a linear system. The goal then reduces to finding the optimal coefficient vector that best represents the time derivative of the given data. We review some recent works on the identification of differential equations. We find some common themes for the improved accuracy: (i) The formulation of linear system with proper denoising is important, (ii) how to utilize sparsity and model selection to find the correct coefficient support needs careful attention, and (iii) there are ways to improve the coefficient recovery. We present an overview and analysis of these approaches about some recent developments on the topic.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

42 extracted references · 37 canonical work pages

  1. [18]

    How much can one learn a partial differential equation from its solution? Foundations of Computational Mathematics , 24(5):1595–1641, 2024

    Yuchen He, Hongkai Zhao, and Yimin Zhong. How much can one learn a partial differential equation from its solution? Foundations of Computational Mathematics , 24(5):1595–1641, 2024. 25

    work page 2024

  2. [1]

    A new look at the statistical model identification

    Hirotugu Akaike. A new look at the statistical model identification. IEEE transactions on automatic control, 19(6):716–723, 1974

    work page 1974

  3. [2]

    Fitting ordinary differential equations to chaotic data

    Ellen Baake, Michael Baake, HG Bock, and KM Briggs. Fitting ordinary differential equations to chaotic data. Physical Review A , 45(8):5524, 1992

    work page 1992

  4. [3]

    A new method for the identification of systems

    Richard Bellman. A new method for the identification of systems. Mathematical Biosciences, 5(1-2):201–204, 1969

    work page 1969

  5. [4]

    Least squares methods.Handbook of numerical analysis , 1:465–652, 1990

    ˚Ake Bj¨ orck. Least squares methods.Handbook of numerical analysis , 1:465–652, 1990. 24

    work page 1990

  6. [5]

    Error analysis of least squares algorithms

    ˚Ake Bj¨ orck. Error analysis of least squares algorithms. In Numerical Linear Algebra, Digital Signal Processing and Parallel Algorithms , pages 41–73. Springer, 1991

    work page 1991

  7. [6]

    Automated reverse engineering of nonlinear dynamical systems

    Josh Bongard and Hod Lipson. Automated reverse engineering of nonlinear dynamical systems. Proceedings of the National Academy of Sciences , 104(24):9943–9948, 2007

    work page 2007

  8. [7]

    Discovering governing equations from data by sparse identification of nonlinear dynamical systems

    Steven L Brunton, Joshua L Proctor, and J Nathan Kutz. Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Proceedings of the national academy of sciences , 113(15):3932–3937, 2016

    work page 2016

Show all 42 references
  1. [8]

    Subspace pursuit for compressive sensing signal reconstruction

    Wei Dai and Olgica Milenkovic. Subspace pursuit for compressive sensing signal reconstruction. IEEE transactions on Information Theory , 55(5):2230–2249, 2009

    2009

  2. [9]

    Uncertainty principles and ideal atomic decomposition

    David L Donoho and Xiaoming Huo. Uncertainty principles and ideal atomic decomposition. IEEE transactions on Information Theory , 47(7):2845–2862, 2001

    2001

  3. [10]

    Coherence pattern–guided compressive sensing with un- resolved grids

    Albert Fannjiang and Wenjing Liao. Coherence pattern–guided compressive sensing with un- resolved grids. SIAM Journal on Imaging Sciences , 5(1):179–202, 2012

    2012

  4. [11]

    On sparse representations in arbitrary redundant bases

    J-J Fuchs. On sparse representations in arbitrary redundant bases. IEEE transactions on Information Theory, 50(6):1341–1344, 2004

    2004

  5. [12]

    Robust and optimal sparse regression for nonlinear pde models

    Daniel R Gurevich, Patrick AK Reinbold, and Roman O Grigoriev. Robust and optimal sparse regression for nonlinear pde models. Chaos: An Interdisciplinary Journal of Nonlinear Science , 29(10):103113, 2019

    2019

  6. [13]

    Uniformly high order accurate essentially non-oscillatory schemes, iii

    Ami Harten, Bjorn Engquist, Stanley Osher, and Sukumar R Chakravarthy. Uniformly high order accurate essentially non-oscillatory schemes, iii. Journal of computational physics , 131(1):3–47, 1997

    1997

  7. [14]

    Group projected subspace pursuit for block sparse signal reconstruction: Convergence analysis and applications

    Roy Y He, Haixia Liu, and Hao Liu. Group projected subspace pursuit for block sparse signal reconstruction: Convergence analysis and applications. Applied and Computational Harmonic Analysis, 75:101726, 2025

    2025

  8. [15]

    Robust identification of differential equations by numerical techniques from a single set of noisy observation

    Yuchen He, Sung-Ha Kang, Wenjing Liao, Hao Liu, and Yingjie Liu. Robust identification of differential equations by numerical techniques from a single set of noisy observation. SIAM Journal on Scientific Computing , 44(3):A1145–A1175, 2022

    2022

  9. [16]

    Group projected subspace pursuit for identification of variable coefficient differential equations (gp-ident)

    Yuchen He, Sung Ha Kang, Wenjing Liao, Hao Liu, and Yingjie Liu. Group projected subspace pursuit for identification of variable coefficient differential equations (gp-ident). Journal of Computational Physics , 494:112526, 2023

    2023

  10. [17]

    Asymptotic the- ory of-regularized pde identification from a single noisy trajectory

    Yuchen He, Namjoon Suh, Xiaoming Huo, Sung Ha Kang, and Yajun Mei. Asymptotic the- ory of-regularized pde identification from a single noisy trajectory. SIAM/ASA Journal on Uncertainty Quantification, 10(3):1012–1036, 2022

    2022

  11. [19]

    The distribution of the flora in the alpine zone

    Paul Jaccard. The distribution of the flora in the alpine zone. 1. New phytologist, 11(2):37–50, 1912

    1912

  12. [20]

    Block subspace pursuit for block-sparse signal reconstruction

    A Kamali, MR Aghabozorgi Sahaf, AM Doost Hooseini, and AA Tadaion. Block subspace pursuit for block-sparse signal reconstruction. Iranian Journal of Science and Technology. Transactions of Electrical Engineering, 37(E1):1, 2013

    2013

  13. [21]

    Ident: Identifying differential equations with numerical time evolution

    Sung Ha Kang, Wenjing Liao, and Yingjie Liu. Ident: Identifying differential equations with numerical time evolution. Journal of Scientific Computing , 87:1–27, 2021

    2021

  14. [22]

    Benchmarking sparse system identification with low-dimensional chaos

    Alan A Kaptanoglu, Lanyue Zhang, Zachary G Nicolaou, Urban Fasel, and Steven L Brunton. Benchmarking sparse system identification with low-dimensional chaos. Nonlinear Dynamics, 111(14):13143–13164, 2023

    2023

  15. [23]

    ¨Uber die beste ann¨ aherung von funktionen einer gegebenen funktionen- klasse

    Andrei Kolmogoroff. ¨Uber die beste ann¨ aherung von funktionen einer gegebenen funktionen- klasse. Annals of Mathematics , 37(1):107–110, 1936

    1936

  16. [24]

    Surfaces generated by moving least squares methods

    Peter Lancaster and Kes Salkauskas. Surfaces generated by moving least squares methods. Mathematics of computation , 37(155):141–158, 1981

    1981

  17. [25]

    Robust low-rank discovery of data- driven partial differential equations

    Jun Li, Gan Sun, Guoshuai Zhao, and H Lehman Li-wei. Robust low-rank discovery of data- driven partial differential equations. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 767–774, 2020

    2020

  18. [26]

    Weak sindy for partial differential equations

    Daniel A Messenger and David M Bortz. Weak sindy for partial differential equations. Journal of Computational Physics , 443:110525, 2021

    2021

  19. [27]

    Weak sindy: Galerkin-based data-driven model selection

    Daniel A Messenger and David M Bortz. Weak sindy: Galerkin-based data-driven model selection. Multiscale Modeling & Simulation , 19(3):1474–1497, 2021

    2021

  20. [28]

    Parameter identification techniques for partial differential equations

    Thomas G M¨ uller and Jens Timmer. Parameter identification techniques for partial differential equations. International Journal of Bifurcation and Chaos , 14(06):2053–2060, 2004

    2004

  21. [29]

    Using noisy or incomplete data to discover models of spatiotemporal dynamics

    Patrick AK Reinbold, Daniel R Gurevich, and Roman O Grigoriev. Using noisy or incomplete data to discover models of spatiotemporal dynamics. Physical Review E, 101(1):010203, 2020

    2020

  22. [30]

    Data-driven iden- tification of parametric partial differential equations

    Samuel Rudy, Alessandro Alla, Steven L Brunton, and J Nathan Kutz. Data-driven iden- tification of parametric partial differential equations. SIAM Journal on Applied Dynamical Systems, 18(2):643–660, 2019

    2019

  23. [31]

    Rudy, Alessandro Alla, Steven L

    Samuel H. Rudy, Alessandro Alla, Steven L. Brunton, and J. Nathan Kutz. Data-driven identification of parametric partial differential equations. SIAM J. Appl. Dyn. Syst. , 18:643– 660, 2019

    2019

  24. [32]

    Data-driven discovery of partial differential equations

    Samuel H Rudy, Steven L Brunton, Joshua L Proctor, and J Nathan Kutz. Data-driven discovery of partial differential equations. Science Advances, 3(4):e1602614, 2017

    2017

  25. [33]

    Learning partial differential equations via data discovery and sparse opti- mization

    Hayden Schaeffer. Learning partial differential equations via data discovery and sparse opti- mization. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sci- ences, 473(2197):20160446, 2017. 26

    2017

  26. [34]

    Distilling free-form natural laws from experimental data

    Michael Schmidt and Hod Lipson. Distilling free-form natural laws from experimental data. science, 324(5923):81–85, 2009

    2009

  27. [35]

    Cross-validation: A review

    Mervyn Stone. Cross-validation: A review. Statistics: A Journal of Theoretical and Applied Statistics, 9(1):127–139, 1978

    1978

  28. [36]

    He, and Hao Liu

    Cheng Tang, Roy Y. He, and Hao Liu. Wg-ident: Weak group identification of pdes with varying coefficients, 2025

    2025

  29. [37]

    Weakident: Weak formulation for identifying differential equation using narrow-fit and trimming

    Mengyi Tang, Wenjing Liao, Rachel Kuske, and Sung Ha Kang. Weakident: Weak formulation for identifying differential equation using narrow-fit and trimming. Journal of Computational Physics, 483:112069, 2023

    2023

  30. [38]

    Fourier features for identifying differential equations (fourierident)

    Mengyi Tang, Hao Liu, Wenjing Liao, and Sung Ha Kang. Fourier features for identifying differential equations (fourierident). arXiv preprint arXiv:2311.16608 , 2023

    2023 arXiv

  31. [39]

    Regression shrinkage and selection via the lasso

    Robert Tibshirani. Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society Series B: Statistical Methodology , 58(1):267–288, 1996

    1996

  32. [40]

    Just relax: Convex programming methods for subset selection and sparse approximation

    Joel A Tropp. Just relax: Convex programming methods for subset selection and sparse approximation. ICES report, 404, 2004

    2004

  33. [41]

    Just relax: Convex programming methods for identifying sparse signals in noise

    Joel A Tropp. Just relax: Convex programming methods for identifying sparse signals in noise. IEEE Transactions on Information Theory , 52(3):1030–1051, 2006

    2006

  34. [42]

    Model selection and estimation in regression with grouped variables

    Ming Yuan and Yi Lin. Model selection and estimation in regression with grouped variables. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , 68(1):49–67, 2006. 27

    2006

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.